Description
TitlePushing the envelope of Wi-Fi networks using distributed multi-user MIMO
Date Created2020
Other Date2020-01 (degree)
Extent1 online resource (xv, 106 pages) : illustrations
DescriptionThis dissertation presents a distributed multi-user MIMO Wi-Fi architecture referred to as D-MIMO that boosts network throughput performance compared to state-of-the-art Wi-Fi access points with co-located antennas. D-MIMO, at a high level, is a technique by which a set of wireless access points are synchronized and grouped together to jointly serve multiple users simultaneously. The cooperation between the access points reduces intra-network interference and hence improves spatial reuse of channels. We study D-MIMO Wi-Fi networks in four broad sections: (i) by prescribing lightweight and effective solutions to the problems of channel access and multi-user MIMO user selection in D-MIMO Wi-Fi, (ii) through experimental evaluations of the proposed solutions on a D-MIMO Wi-Fi network implemented in an indoor testbed using software defined radio platforms, (iii) by constructing a deep reinforcement learning framework to address dynamic resource management in D-MIMO Wi-Fi networks, and (iv) by investigating the benefits that the D-MIMO architecture brings to dense Wi-Fi networks operating in mmWave (60 GHz) bands. These components form the original contributions of this dissertation to knowledge.
Designing a D-MIMO Wi-Fi network invites us to revisit fundamental Wi-Fi concepts such as carrier sensing multiple access that governs medium/channel access among Wi-Fi access points. We propose a medium access protocol for D-MIMO that assimilates channel sensing observations from different access points to resolve channel contention among D-MIMO groups. We also propose a novel way of using channel reciprocity and the network topology to select downlink multi-user (MU) MIMO recipients without requesting any form of channel state information feedback from the users during the selection phase. The proposed solutions are lightweight, do not require modifications at the user equipment, and hence will work with legacy 802.11ac devices. We compare the performance of the D-MIMO configuration to that of baseline dense Wi-Fi deployments (access points with co-located antennas), operating in 5 GHz bands, through extensive network simulations. We observe an improvement of 3.5x in median and 191% in mean user throughput, as well as a reduction of 61% in channel access delay with D-MIMO.
Next, we present an implementation of a distributed MIMO Wi-Fi group---using software defined radio platforms---in an indoor experimental testbed. The implemented setup consists of four two-antenna Wi-Fi access points (synchronized in time and phase using a GPS-disciplined clock reference system) and twenty two-antenna users, and is compliant with the 802.11ac very high throughput framework. We use this setup to serve as a proof-of-concept of the proposed lightweight MU-MIMO user selection algorithm. Through extensive experimental evaluations, we demonstrate that the proposed algorithm outperforms a simple random user selection strategy by achieving an improvement of up to 60% in median and 43% in mean group throughput performance. Furthermore, the proposed user selection algorithm performs close to optimality---the difference in performance between the proposed user selection algorithm and optimal user selection is a mere 13%.
As the third installment of this dissertation, we address two dynamic resource management problems germane to D-MIMO Wi-Fi networks: (i) channel assignment of D-MIMO groups, and (ii) deciding how to cluster access points to form D-MIMO groups, in order to maximize user throughput performance. These problems are known to be NP-Hard for which only heuristic solutions exist in literature and we explore the potential of harnessing principles from deep reinforcement learning (DRL) to address these challenges. We construct a DRL framework through which a learning agent interacts with a D-MIMO Wi-Fi network, learns about the network environment, and successfully converges to policies that effectively address the aforementioned challenges. Through extensive simulations and on-line training based on D-MIMO Wi-Fi networks, we demonstrate the efficacy of DRL agents in achieving an improvement of 20% in user throughput performance compared to heuristic solutions, particularly when network conditions are dynamic. This work also showcases the effectiveness of DRL agents in meeting multiple network objectives simultaneously, for instance, maximizing throughput of users as well as fairness of throughput distribution among them.
In the final part of this dissertation, we consider dense Wi-Fi networks operating in mmWave (60 GHz) bands and use the D-MIMO architecture to improve user throughput performance in these networks compared to baseline arrangements. Rigorous network simulation results reveal an enhancement of 395% in average user throughput and a reduction of 75% in channel access delay with D-MIMO compared to baseline. We observe an interesting behavior wherein a user achieves very high modulation and coding scheme indices more number of times with the baseline configuration compared to D-MIMO, especially when the user is located close to an access point (AP). This behavior can be ascribed to two causes: i) a higher probability of line-of-sight of the short distance AP-user link (that favors baseline), and ii) a ramification of the use of zero-forcing precoding to cancel inter-user interference in D-MIMO. This observation motivates the design of future networks as amalgams of both baseline and D-MIMO arrangements.
NotePh.D.
NoteIncludes bibliographical references
Genretheses, ETD doctoral
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile